The Lost Cycles Toolkit, which automates the process of generating models under the LCA approach, is an example of the importance of auxiliary tools and techniques in understanding the performance of parallel programs. The toolkit automates experimental design, performance measurement, univariate modeling, and multivariate modeling. It also provides feedback to the user on the quality of the resulting model, and allows the user to trace inaccuracies back to the component models and performance data. This potential for iterative model refinement means that the tools can aggressively automate processes like model selection, since the user is alerted whenever a tool's decision or assumptions are inaccurate. The tool has been shown to be a valuable resource, helping the user in all steps of the modeling process. Thus, it makes performance modeling easier and accessible for parallel programmers that do not have modeling and experimental design skills.
The LCT implementation is a good start for the incorporation of modeling techniques within the Carnival framework, since it is already part of the necessary infrastructure for model generation and understanding, and requires only a few enhancements to be compatible with Carnival .